kasetsart journal natural science

234
categories including puffed snacks, potato chips,
dried squid, popcorn, fish snacks, nuts, and prawn
crackers. Before starting the first section, all
panelists were advised how to evaluate samples.
The 9-point hedonic scale was used for preference
ratings (appearance, aroma, taste, texture, and
overall liking), and there was a 10 minute break
between each section.
Data were then analyzed by SPSS 10.0.
Cluster analysis was used to classify respondents
by their overall liking scores for each product
category. Overall liking of each respondent group
for each product category was estimated from their
attribute liking scores by multiple regression
(stepwise) as the equation below.
Overall liking = A (Appearance liking) + B
(Aroma liking) + C (Taste
liking) + D (Texture liking)
1.2 Quantitative descriptive analysis (QDA)
data
Thirty semi-trained panelists were used
to obtain QDA data for the same 21 snack products.
They were selected and trained by the modified
methods of ISO3972:1991 and ISO4120:1983,
respectively.
2. Quality function deployment
A HOQ was built for each snack category
to translate the attribute liking into QDA attributes.
The importance scores of each attribute liking were
considered from its coefficient in the multiple
regression between the attribute liking scores and
the overall liking scores.
The impact between each pair of attribute
liking and related QDA attribute, and the
relationship between each pair of QDA attributes
were considered by Pearson correlation. Cohen
(1995) suggested that impacts in the middle of
HOQ should be considered into 3 levels consisting
of possibly related which was valued as 1,
moderately related which was valued as 3, and
Kasetsart J. (Nat. Sci.) 40(1)
strongly related which was valued as 9. In the
study, therefore, if a correlation coefficient was ≥
0.9 or was an inverted U curve, then it was defined
as being strongly related and scored as 9 for the
impact of attribute liking and related QDA attribute
(in the middle of house), and as + + or -- for the
relationship between a pair of QDA attributes (on
the roof). 0.6 ≤ correlation coefficient < 0.9 was
defined as being moderately related and scored as
3 and as + or –, respectively. In case of 0.4 ≤
correlation coefficient < 0.6 it was defined as
being slightly related and scored as 1. The
contribution (priority) for each QDA attribute was
then computed by summing the multiplication of
impact and importance scores.
The direction of goodness for each QDA
attribute was considered from the sign + and – of
the correlation coefficient; + meant high intensity
is good and vice versa in case of minus. For the
inverted U curve, it meant the intensity of that
sensory attribute should be optimized.
3. Reverse engineering
3.1 Creating a set of equations
The preference scores and QDA data
were transformed to z-scores. The 21 products
were then mapped in a coordinate system by
reducing their QDA data with principal component
analysis (PCA). The products’ PC scores and their
square terms were used as independent variables
in the equations for estimating the degree of
preference for each of the sensory attribute as well
as the overall liking, and the equations for
evaluating the intensity of the QDA attributes by
multiple regression (stepwise).
3.2 Finding the target
The overall liking equation was
maximized for each product category by using the
solver command in Excel 2000. These maximized
PC scores were then used in the QDA attribute
equations to get the target product profiles.